Selected abstracts from the Journal of the Experimental Analysis of Behavior, November 1993.

Three individuals with intellectual disability participated in two experiments on arbitrary conditional discriminations — the kind of matching task where the learner must pick a specific comparison stimulus whenever a particular sample appears, with no physical similarity connecting them.

In Experiment 1, each participant worked on two conditional discriminations per day: one under trial-and-error conditions, and one under a component training procedure. The component procedure first established the comparison discrimination and its rapid reversal, then used differential naming to build a successive discrimination between the sample stimuli. Sample naming was maintained as block size was gradually decreased across sessions until presentation was fully randomized — matching the structure of trial-and-error training but with the naming response intact.

Experiment 2 then systematically withdrew elements of the component procedure across successive conditional discriminations to test whether a learning-set effect had developed.

Two of the three subjects learned only under component training in Experiment 1; the third was inconsistent across conditional discriminations. One of the successful subjects subsequently learned rapidly and consistently under trial-and-error procedures alone.

In Experiment 2, one subject ultimately nearly always learned under trial-and-error conditions, and the other learned under trial-and-error combined with differential sample naming. This learning-set effect — acquiring new conditional sets in fewer trials over time — indicates the component package trained the underlying process, not just the specific stimulus pairs.

[citation removed] later applied the same logic with autistic preschoolers using computer-assisted multiple-exemplar training, demonstrating that the principle of building stimulus control through differential naming can be adapted to younger populations and technology-based delivery.

[citation removed] extended the approach to Kanji reading with students with developmental disabilities, using nameable stimuli to build conditional relations that maintained over time — consistent with the learning-set outcomes reported here.

[citation removed] showed that simpler feature-detection discriminations can be taught with prompting and fading alone, without a naming component. The contrast clarifies when naming adds value: it helps most when the task requires a conditional 'if A then pick B' relation, not just detection of a single feature.

When trial-and-error stalls on a conditional discrimination task, the data here support two targeted additions: have the learner name the sample aloud, and present trials in decreasing block sizes before randomizing. The naming step sharpens stimulus control over the sample, and the gradual fade gives errors nowhere to hide. One subject in this study went on to master new sets under plain trial-and-error — the clearest sign that the component procedure can build a durable learning process, not just performance on a single task.